Post-stress bacterial cell death mediated by reactiveoxygen speciesYuzhi Honga, Jie Zengb, Xiuhong Wanga,c, Karl Drlicaa,d, and Xilin Zhaoa,b,d,1

aPublic Health Research Institute Center, New Jersey Medical School, Rutgers Biomedical and Health Sciences, Rutgers University, Newark, NJ 07103; bStateKey Laboratory of Molecular Vaccinology and Molecular Diagnostics, School of Public Health, Xiamen University, Xiamen, Fujian Province 361102, China;cDepartment of Biochemistry and Molecular Biology, Harbin Medical University, Harbin, Heilongjiang Province 150081, China; and dDepartment ofMicrobiology, Biochemistry & Molecular Genetics, New Jersey Medical School, Rutgers Biomedical and Health Sciences, Rutgers University, Newark, NJ 07103

Edited by Carl F. Nathan, Weill Medical College of Cornell University, New York, NY, and approved March 14, 2019 (received for review January 31, 2019)

Antimicrobial efficacy, which is central to many aspects of medicine,is being rapidly eroded by bacterial resistance. Since new resistancecan be induced by antimicrobial action, highly lethal agents thatrapidly reduce bacterial burden during infection should help restrictthe emergence of resistance. To improve lethal activity, recent workhas focused on toxic reactive oxygen species (ROS) as part of thebactericidal activity of diverse antimicrobials. We report that whenEscherichia coliwas subjected to antimicrobial stress and the stressorwas subsequently removed, both ROS accumulation and cell deathcontinued to occur. Blocking ROS accumulation by exogenous miti-gating agents slowed or inhibited poststressor death. Similar resultswere obtained with a temperature-sensitive mutational inhibition ofDNA replication. Thus, bacteria exposed to lethal stressors may notdie during treatment, as has long been thought; instead, death canoccur after plating on drug-free agar due to poststress ROS-mediatedtoxicity. Examples are described in which (i) primary stress-mediateddamage was insufficient to kill bacteria due to repair; (ii) ROS over-came repair (i.e., protection from anti-ROS agents was reduced byrepair deficiencies); and (iii) killing was reduced by anti-oxidativestress genes acting before stress exposure. Enzymatic suppressionof poststress ROS-mediated lethality by exogenous catalase supportsa causal rather than a coincidental role for ROS in stress-mediatedlethality, thereby countering challenges to ROS involvement in anti-microbial killing. We conclude that for a variety of stressors, lethalaction derives, at least in part, from stimulation of a self-amplifyingaccumulation of ROS that overwhelms the repair of primary damage.

reactive oxygen species | poststress cellular response | antimicrobial |antioxidant | damage repair

Discovering ways to manage antimicrobial resistance is amongthe most important medical challenges of our time (1).

Since many antimicrobials can stimulate the production of re-sistant mutants, often via the SOS response (2–8), one way tolimit the emergence of new resistance is to more rapidly andextensively reduce pathogen populations during infection. To-ward that end, we and others have been studying how antimi-crobials and other lethal stressors kill bacteria. Recent work hasdrawn attention to the contribution of stress-stimulated accu-mulation of toxic reactive oxygen species (ROS) (9–21). Findingways to stimulate ROS-mediated killing could in principle en-hance the efficacy of a broad range of antimicrobials. However,an ROS contribution to antimicrobial killing became contro-versial when the original observation (9) was challenged (22–24).Subsequent work countered many of the challenges (13–15, 25)and extended the phenomenon to thymineless death (26), phageinfection (27), the type VI secretion system (27), and over-expression of a MalE-LacZ fusion (28). Moreover, nitric oxideand hydrogen sulfide interfere with antimicrobial killing by sup-pressing ROS generation/accumulation (17–19). Among thequestions to be addressed are whether ROS accumulation is acause or a consequence of cell death, and whether intracellularROS levels are sufficient to kill bacteria once the original inducingstressor is removed, that is, whether ROS accumulation can beself-sustaining or even self-amplifying.

ROS, which are commonly considered to include superoxide,hydrogen peroxide, and hydroxyl radical, cause several types ofintracellular damage. For example, hydroxyl radical breaks DNA,peroxidates lipids, and carbonylates proteins (14). ROS also oxi-dize the dGTP and dCTP pools, causing misincorporation of basesinto DNA that leads to double-stranded breaks (DSBs) in DNAthrough disrupted repair intermediates (11, 29, 30). In principle,ROS-mediated damage, which is secondary to the primary stress-induced lesion, could stimulate additional rounds of ROS accu-mulation, thereby making ROS accumulation a self-amplifying,unstoppable process that could be the terminal step in a bacte-rial response to lethal stress. Whether endogenous ROS are sufficientto kill cells is unclear, because the lethal effect from the primarystressor and that from subsequent ROS accumulation have not beenseparated. Indeed, it has been suggested that endogenous ROSconcentrations are unlikely to reach lethal levels (24).In the present work, we devised a way to separate lethality due to

a primary stressor from that triggered by subsequent intracellularROS accumulation. In this assay, we treated cultured Escherichiacoli with a lethal stressor, removed the stressor, and plated thebacteria on stressor-free agar containing an agent known to suppressthe accumulation of toxic ROS. We found that ROS-mitigatingagents, present in or on agar, interfered with killing by diverse le-thal stressors, even after stressor removal. In these cases, ROS is acause, not a consequence, of stress-induced cell death. In addition, we

Significance

Death, one of the most important events for all organisms, isthought to derive from a variety of lethal assaults. The discoveryof a general, reactive oxygen species (ROS)-dependent mecha-nism that contributes to killing by diverse stressors was surpris-ing. The finding that ROS levels continue to surge and killbacteria even after removal of the initiating stressor adds a newdimension to stress-mediated lethality, demonstrating that theROS surge is self-amplifying. Once an ROS threshold is exceeded,the death process becomes self-driven (i.e., self-sustainable). Inaddition to providing a new way to think about the response ofbacterial populations to lethal stress, our findings encourageefforts to target oxidative stress pathways as adjuncts of anti-microbial therapy that will increase lethality and help restrict theemergence of resistance.

Author contributions: Y.H., K.D., and X.Z. designed research; Y.H., J.Z., and X.W. per-formed research; K.D. and X.Z. contributed new reagents/analytic tools; Y.H., K.D., andX.Z. analyzed data; and Y.H., K.D., and X.Z. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Published under the PNAS license.

Data deposition: The RNA sequencing data have been deposited in the Gene ExpressionOmnibus (GEO) database, https://www.ncbi.nlm.nih.gov/geo (accession no. GSE114262).1To whom correspondence should be addressed. Email: zhaox5@njms.rutgers.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1901730116/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1901730116 PNAS Latest Articles | 1 of 8

MICRO

BIOLO

GY

Dow

nloa

ded

by g

uest

on

June

12,

202

0

report a case in which primary damage to DNA was insufficient tocause cell death; accumulation of toxic ROS is required unless cellscontain a deficiency in primary damage repair. The work also revealsa misconception concerning the traditional antimicrobial killing assayin which bacteria are thought to be dead at the time of posttreatmentplating; they can actually still be alive, dying on drug-free agar platesduring poststress recovery due to ROS accumulation. Overall, ourfindings provide insight into antimicrobial lethality and support ef-forts to find broad-spectrumROS-related enhancers of antimicrobialaction.

ResultsPoststress Cell Death Associated with Antimicrobial Stress. To sep-arate lethality due to a primary stressor from that due to the ac-cumulation of ROS, we examined ROS-mediated death occurringafter removal of the primary stressor (Fig. 1A). For example, wetreated E. coli cultures with nalidixic acid [∼1 minimum inhibitoryconcentration (MIC); SI Appendix, Table S1], washed cells toremove the drug, and then plated cells on agar containing a sub-inhibitory concentration (0.5 MIC) of bipyridyl to halt the pro-duction and accumulation of hydroxyl radical. The presence ofbipyridyl in agar raised survival by ∼30-fold (Fig. 1B), indicatingthat ROS are responsible for death (SI Appendix, Fig. S1 A‒G).Residual nalidixic acid is unlikely to contribute to the ROS-mediated killing on agar, because quinolone-gyrase-DNA com-plexes are reversible, and washing cells to remove quinolone fullyrestores DNA synthesis blocked by formation of the complexes

(31–33); if the complexes are rendered irreversible by crosslinkingof quinolone to gyrase, DNA synthesis is not restored (32). Inaddition, a mass spectrometry-based assay showed that washingand diluting cells lowered intracellular quinolone concentration to0.002 times MIC (SI Appendix, Fig. S1H). Since MIC correlateswith cleaved-complex formation (34, 35), which results in the pri-mary quinolone-mediated lesion, few complexes are likely to bepresent at 0.002 MIC. Moreover, no growth inhibition was ob-served at nalidixic acid concentrations as high as 0.5 MIC (SI Ap-pendix, Fig. S1I). These data are consistent with biochemical studiesshowing that dilution reverses quinolone-topoisomerase-DNAcomplex formation (36). Finally, potential growth-inhibitory ef-fects of anti-ROS agents on stressor action are irrelevant, due tostressor removal. We conclude that poststress ROS-mediated tox-icity is responsible for lethality due to nalidixic acid treatment.In a similar experiment, E. coli cells were treated with tri-

methoprim to induce a process similar to thymineless death thatis also associated with ROS accumulation (SI Appendix, Fig. S2A–D) (26, 37). As with nalidixic acid, trimethoprim (6 MIC) wasremoved before plating by washing and diluting cells. Whenapplied to LB agar, the samples were overlaid with catalase todegrade peroxide and thereby remove a source of hydroxyl radical.Catalase increased survival by 5- to 25-fold (Fig. 1C and SI Ap-pendix, Fig. S2E). Mass spectrometry measurement of intracellulardrug concentration indicated that washing/dilution of trimethoprim-treated cells removed most of the drug, lowering the drug con-centration to 0.06 MIC (SI Appendix, Fig. S2F). Thus, trimethoprim

A Log-phase cells

Agar

Stress ROS

Death

(i)

Remove stressor

Survival

CatalaseThioureaBipyridyl

(iii)

(ii)

Agar

(iv)

F

catal

ase +

0 h

catal

ase +

48 h

catal

ase +

40 h

catal

ase +

30 h

catal

ase +

24 h

catal

ase +

16 h

catal

ase +

6 h

catal

ase +

2 h

No cata

lase

Input

102

100

G

0 1 2 3Time (h)

Agar + thiourea % In

put c

ells

reco

vere

d

Agar + catalase

102

100

10-2

Agar % In

put c

ells

reco

vere

d

B

AgarAgar + Bip

0 1 2 3Time (h)

4

100

102

0 1 2Time (h)

1.5

C

102

100

D102

101

10-2

0 1 2 3Time (h)

Agar Agar + BTS%

Inpu

t cel

ls re

cove

red

% In

put c

ells

reco

vere

d

% In

put c

ells

reco

vere

d

Nalidixic acid Ampicillin + BT

dnaB-Ts dnaB-TsNal 2.5 h → Agar 60 min

No drug → Agar 60 min

Nal 2.5 h → Agar 0 min

5 μm

E

10-1ROS

ROS

Agar Agar + catalase

Trimethoprim

Fig. 1. ROS-mediated poststress death (PSD). (A) Assay design. Cultures were stressed, the stressor was removed, and cells were plated on agar containing/lackingagents that interfere with ROS accumulation; then cfu was determined. (B) Nalidixic acid-stimulated PSD. Wild-type strain 3001, treated with drug (1 MIC), wasplated on agar containing/lacking bipyridyl (Bip, 0.5 MIC; n = 5). (C) Trimethoprim-stimulated PSD. Wild-type strain (2428), treated with trimethoprim (6 MIC), wasplated on agar overlaid (or not) with catalase (n = 4). (D) Ampicillin-stimulated PSD. As in B but with ampicillin (2.5 MIC) in the presence/absence of bipyridyl (0.4MIC) plus thiourea (0.45 MIC) (BT). Cultures were plated on agar containing/lacking the bipyridyl-thiourea combination plus sucrose (BTS; n = 3). (E) ROS accu-mulation on agar after removal of nalidixic acid (2.5 h treatment) and pulse-labeled with carboxy-H2DCFDA (20 min) immediately after drug removal. n = 4. (F)Inhibition of dnaB-Ts–stimulated PSD. Strain 2429 was shifted to 42 °C for the indicated times and then plated at 30 °C on agar, agar plus 15 mM (0.15 MIC)thiourea (n = 5), or agar in which spotted samples were overlaid with catalase (200 U; n = 3). The percentage of cells recovered is relative to cfu at the point atwhich cultures were shifted to 42 °C. (G) Time for PSD completion. dnaB-Ts mutant was incubated at 42 °C for 2.5 h and then plated at 30 °C for the indicatedtimes before catalase was added for 48 h (n = 6). n, number of independent experiments. Data are mean ± SD.

2 of 8 | www.pnas.org/cgi/doi/10.1073/pnas.1901730116 Hong et al.

Dow

nloa

ded

by g

uest

on

June

12,

202

0

action is a second example of poststress ROS-mediated death. Theprotective effect of catalase also provides strong evidence that ROScause death rather than resulting from it.We next examined the lethal action of ampicillin, which in-

hibits cell wall synthesis and can lead to cell lysis under someconditions (38). Previous work showed that bipyridyl or thiourea,added to broth cultures during ampicillin treatment, inhibitskilling (9, 10), a finding consistent with increased ROS accu-mulation during ampicillin treatment (SI Appendix, Fig. S2 Gand H). To prevent ROS from accumulating so rapidly that cellsare killed in broth culture, an event that would obscure detectionof poststress death on agar (see Discussion), we added sub-inhibitory concentrations of bipyridyl plus thiourea to brothcultures containing ampicillin at 2.5 MIC. Subsequent plating ondrug-free agar after removal of ampicillin from cells showed thatthiourea plus bipyridyl in agar increased survival by 350-fold (Fig.1D and SI Appendix, Fig. S2 I and J). Ampicillin was readily re-moved from cells by the washing/dilution procedure, with massspectrometry measurements indicating 600- and 20,000-fold lowerdrug concentrations in cells after a single washing compared withthose seen for trimethoprim and nalidixic acid, respectively. Afterthe four 10-fold dilutions used with the poststress killing assay, thelevel of ampicillin was below the detection limit (a decrease of atleast 100-fold relative to cells that were not diluted) (SI Appendix,Fig. S2K).For the three antimicrobials tested above, we looked for the

accumulation of ROS in cells on agar. Immediately after removalof antimicrobial, we pulse-labeled the culture for 20 min withcarboxy-H2DCFDA, an ROS-sensitive fluorescent dye, plated thecells, and examined them by fluorescence microscopy. Little fluo-rescence was evident when cells were examined immediately afterlabeling with the dye, indicating that little ROS was carried overwhen cells were plated on agar. However, after incubation on agarfor 1 h, a strong ROS signal was seen in cells that had been treatedwith nalidixic acid, trimethoprim, or ampicillin (Fig. 1E and SIAppendix, Fig. S3). These data indicate that damage due to theaction of lethal antimicrobials stimulates a self-amplifying accu-mulation of ROS that continues after antimicrobial removal.

Poststress Cell Death with a dnaB-Ts Mutant. To extend the study toa conditional lethal mutation (a nonantimicrobial stress), weexamined a temperature-sensitive helicase mutant (dnaB-Ts; SIAppendix, Fig. S4A). An ROS increase was observed as elevatedfluorescence of ROS-indicator dyes (carboxyl-H2DCFDA andPeroxy Orange 1) and as increased activity of the ROS-sensitivepromoters of oxyS and soxS (13) (SI Appendix, Fig. S4 B–E). Asexpected, ROS accumulation paralleled a drop in survival (SIAppendix, Fig. S4F). Moreover, a shift to restrictive temperaturefor 1 h caused a twofold increase in expression of four ROS-responsive detoxifying genes (SI Appendix, Fig. S4G). The ad-dition of a subinhibitory combination of bipyridyl plus thioureasuppressed the ROS accumulation and completely blocked kill-ing (SI Appendix, Fig. S4 H–J). Control experiments showed thatthis combination of bipyridyl plus thiourea had little effect ongrowth rate (SI Appendix, Fig. S4K). Moreover, no ROS accu-mulation was detected in the dnaB-Ts mutant grown at permis-sive temperature or in wild-type cultures incubated at 42 °C;likewise, no autofluorescence was observed with the mutant at42 °C (SI Appendix, Fig. S4 L–P). Thus, the temperature shiftstimulated lethal events associated with toxic ROS accumulation.The dnaB-Ts system appeared to be suitable for assaying post-

stress death following a pulse of incubation at 42 °C, becausedownshift from 42 °C to permissive temperature (30 °C) allowedrecovery of DNA synthesis in 1 h (SI Appendix, Fig. S5A). Sincemutant and wild-type cells exhibited the same growth rate at 30 °C,and since the mutant showed similar cfu on agar incubated at 30 °Cand 22 °C (SI Appendix, Fig. S5 B and C), the activity of mutant

DnaB appears to be normal at permissive temperature, as is re-quired for a downshift to eliminate stress.When we exposed dnaB-Ts cultures to 42 °C for 2.5 h, fol-

lowed by immediate chilling and plating at 30 °C, survival wasonly 0.2% relative to cell density at the time when cultures wereshifted to 42 °C (Fig. 1F). However, survival was almost 100%when mutant cells incubated at 42 °C were subsequently plated at30 °C on agar either containing thiourea or overlaid with catalaseand then incubated at 30 °C (Fig. 1F). In control experiments, wefound that survival of mutant cells maintained at 30 °C throughoutthe experiment was unaffected by either thiourea or catalase onagar; likewise, substitution of an unrelated protein (BSA) for cat-alase had no effect on growth/survival of the mutant incubated atrestrictive temperature before plating (SI Appendix, Fig. S5 D–F).The addition of either thiourea or catalase to agar also blockedkilling of a dnaB-Ts-ΔkatG hyperlethal double mutant (SI Ap-pendix, Fig. S4H), indicating that the anti-ROS treatments wereeffective even when the absence of the KatG catalase allowed el-evated levels of oxidative stress (20, 26). Overall, poststress ROSaction accounts for DnaB-Ts–mediated killing.Poststress death at 30 °C, stimulated by a 2.5-h treatment of the

dnaB-Ts mutant at 42 °C, takes several hours to complete; evenafter incubation on agar at 30 °C for 2–6 h, the addition of catalaseprevented the death of most cells (Fig. 1G). However, when cat-alase addition was delayed for longer than 40 h, little protectionwas observed (Fig. 1G). Consequently, by 40 h, the poststress deathprocess was largely complete; catalase could not revive dead cells.These data show that ROS cause death, because if only bacterialgrowth were blocked on agar or if cells entered a viable but notculturable physiological state, then catalase-mediated rescue wouldhave been independent of the time of catalase addition.

Genes Involved in dnaB-Ts–Triggered Poststress Cell Death. Sincepoststress ROS contributes 100% to the death of the dnaB-Tsmutant (Fig. 1F and SI Appendix, Fig. S4H), we used the mutantto search for genes whose deficiency may affect poststress death. Avariety of DNA repair mutations showed little effect with the dnaB-Ts mutant (SI Appendix, Fig. S6 A–C), but a deficiency in ahpC/F,which encode a peroxidase (39), almost completely suppressed thekilling arising from incubation at restrictive temperature (Fig. 2A).The ahpC deficiency had no effect on culture growth of the dnaB-Tsmutant. An ahpF deficiency showed a protective behavior similar tothat seen with ahpC (SI Appendix, Fig. S7 A and B). The results withahpC and ahpF deficiencies were paradoxical, because reducedROS detoxification should increase rather than decrease killing.To address the ahpC/F paradox, we performed RNA-seq with

the dnaB-Ts and dnaB-Ts–ΔahpC mutants (40). The combinationof the two mutations elevated the expression of many genes in-volved in the reduction of oxidative stress; elevation occurred atboth permissive and nonpermissive temperatures (Fig. 2B and SIAppendix, Table S2). Elevated expression of protective genes atpermissive temperature could enable cells to better tolerate dam-age from subsequent stress-stimulated ROS increases. Indeed, theahpC mutation reduced ROS accumulation at restrictive temper-ature when present in the dnaB-Ts mutant (Fig. 2C); it alsoeliminated killing in a dnaB-Ts strain that carried a deficiency inkatG (Fig. 2D), thereby overcoming the loss of KatG-mediatedhyperlethality. Moreover, the dnaB-Ts–ΔahpC mutant grown atpermissive temperature exhibited protection from killing by β-lactamsand quinolones (SI Appendix, Fig. S7 C and D). Collectively, theseobservations explain the protective action of the ahpC deficiency, asincreased expression of protective genes and support the conclusionthat ROS contribute in a general way to antimicrobial lethality.

Contribution of Primary Damage and Damage Repair to Cell Death.To better understand the roles of ROS and primary damage instress-mediated killing, we used RecN-YFP–based fluorescencemicroscopy (26, 41) to monitor the accumulation of DSBs in DNA;

Hong et al. PNAS Latest Articles | 3 of 8

MICRO

BIOLO

GY

Dow

nloa

ded

by g

uest

on

June

12,

202

0

both quinolone and ROS are known to generate DNA breaks (11,26, 28, 30, 42, 43). Treatment with the quinolone nalidixic acid ledto rapid accumulation of DSBs; after 1 h treatment, 97% of cellsscored positive (Fig. 3A). Agents that suppress ROS accumulation(thiourea plus bipyridyl) failed to block DSB accumulation butinhibited killing (SI Appendix, Fig. S1 E and F). These data indicatethat (i) quinolones can generate DSBs without ROS participationand (ii) primary quinolone-mediated DSBs are insufficient to killcells. Finding examples in which the primary damage (e.g., nalidixicacid-mediated DSBs) fails to correlate with death while ROS ac-cumulation does correlate with death constitutes evidence for adecisive role of ROS in stress-mediated killing.The behavior of DNA breaks associated with the dnaB-Tsmutant

differed from that described for nalidixic acid. When the dnaB-Tsmutant was shifted to 42 °C, DSBs appeared more slowly than seenwith nalidixic acid, beginning at 1 h after upshifting and being seenin almost every cell by 3 h after upshifting (Fig. 3C). Moreover,dnaB-Ts–mediated DSBs were not observed when ROS accumu-lation was suppressed in broth cultures by bipyridyl plus thiourea orwhen cells contained both the ΔahpC and dnaB-Ts mutations (Figs.2C and 3 D and E and SI Appendix, Fig. S4I). However, as withnalidixic acid, dnaB-Ts–mediated DSBs were insufficient to killcells, since 100% of the cells in which 60% contained DSBs wereviable at 2 h after the temperature upshift (Figs. 1F and 3C). Thesedata suggest that incubation at nonpermissive temperature leads toROS-induced DSBs, but that those initial lesions do not kill cellswithout additional ROS-mediated damage (Figs. 1F and 2A).DNA repair is one explanation for the failure of the quinolone-

induced DNA DSBs to kill cells when an ROS surge is blocked.With cells harboring a deficiency in recA, treatment with a bipyridyl-thiourea combination failed to fully protect cells from quinolone-mediated killing (Fig. 4 A and B and SI Appendix, Fig. S6 D and E),in contrast to the full protection seen with wild-type cells. More-over, a combination of recA and recBC deficiencies further reducedprotection by a bipyridyl-thiourea combination. These observationsindicate that successive reduction in DNA repair capacity may in-crementally lower the requirement for ROS to execute quinolone-mediated death. Examination of mutants deficient in repair ofprotein damage indicated that for kanamycin, unrepaired primarydamage also reduces the contribution of ROS to death (Fig. 4 C

and D). Thus, in repair-proficient strains, death may derive fromROS overwhelming repair.To assess the nature of ROS-mediated secondary damage dur-

ing stress, we treated E. coli cultures with oxolinic acid, which, likenalidixic acid, is a first-generation quinolone that generates DNAlesions (34) as primary damage. We then measured protein car-bonylation and lipid peroxidation, neither of which was expected tobe a direct effect of the drug (e.g., ROS-mediated secondary dam-age), and both were increased unless thiourea and bipyridyl werepresent (SI Appendix, Fig. S8 A–C). Likewise, kanamycin, whichcreates mistranslated/misfolded proteins (44) as primary damage,caused lipid peroxidation; polymyxin, which primarily acts onlipids (45), produced protein carbonylation, and ampicillin, whichprimarily blocks wall synthesis (38), created ROS-dependent DNAdamage (SI Appendix, Fig. S8 D‒G). In principle, such secondarydamage can stimulate additional rounds of ROS production thatgenerate further (tertiary) damage. As expected, treatment of cellswith oxolinic acid generated more lipid peroxidation in a proteinrepair mutant than in wild-type cells (SI Appendix, Fig. S8C).These data support the conclusion that the ROS surge can be self-amplifying; successive rounds of ROS attack occur and overwhelmcellular repair with massive secondary and tertiary damage.

Exogenous Hydrogen Peroxide Stimulates Cell Death During Stress.To examine the converse of suppressing ROS accumulation withantioxidants, we monitored the effect of hydrogen peroxide addedto cells also stressed by other agents. We first determined a sub-lethal dose (1.5 mM, which in a 3-h treatment did not kill other-wise unstressed cells; SI Appendix, Fig. S9 A and B). We nextadded sublethal peroxide for 20 min either (i) after shifting thednaB-Ts mutant to 42 °C for 2.5 h and then to 30 °C or (ii) after a2-h treatment with nalidixic acid (∼1 MIC). With both stressors,exogenous peroxide added as a sublethal dose only to broth cul-tures eliminated protection by bipyridyl subsequently added toagar following removal of the stressor (Fig. 5 A and B). Exogenoushydrogen peroxide also contributed to ampicillin-mediated death.When bipyridyl and thiourea were added to broth cultures, theyinhibited cell lysis (Fig. 5C); subsequent addition of sublethal(1.5 mM) hydrogen peroxide rapidly lysed otherwise intact cells(Fig. 5D). High concentrations of peroxide alone also lysed cells

A102

% In

put c

ells

reco

vere

d

0 1 2 3Time (h)

dnaB-Ts dnaB-Ts ahpC

dnaB-Ts ahpC + catalase

dnaB-Ts ahpCdnaB-Ts katG

dnaB-Ts katGahpC0 1 2 3

D

102 103 104

Fluorescence

Cel

l num

bers

(x 1

00)

0

6

C

ahpC_0 h

ahpC_3 h

0 h

3 h

B

100

10-2

102

100

10-2

10-4

Time (h)

Up-regulatedFold (42 oC)Fold (30 oC)

% In

put c

ells

reco

vere

d

0 1 2 3Time (h)

dnaB-Ts + catalase

dnaB-Ts

dnaB-Ts

dnaB-Ts ahpC + thioureadnaB-Ts + thiourea

dnaB-Ts ∆ahpC dnaB-Tsvs

grxA dps katG soxS ahpF trxC ydbK flu yaaA hemH fpr entA entB entC entE sufA sufB sufC sufD sufS cirA fumC25.2 12.5 9.6 8.7 4.5 3.5 2.6 2.0 2.2 2.2 1.9 1.4 1.6 1.4 1.4 1.8 1.8 1.6 1.5 1.5 1.7 1.735.7 5.1 11.4 4.8 4.6 2.7 3.2 2.8 1.7 1.7 2.1 1.9 2.2 2.1 2.3 2.3 2.5 2.5 2.5 2.1 2.0 2.0

Fig. 2. Effect of ahpC deficiency on dnaB-Ts–stimulated poststress death. (A) Inhibition of dnaB-Ts–stimulated death. Strains 2429 (dnaB-Ts) and 3780 (dnaB-TsΔahpC) were shifted to 42 °C for the indicated times and then plated at 30 °C on agar, agar plus catalase (Left), or agar plus 15 mM (0.15 MIC) thiourea (Right). Afterincubation, cfu were measured (n = 3). (B) Up-regulation of antioxidant and stress response genes associated with deficiency of ahpC in a dnaB-Ts strain (strains as inA). RNA-seq data are from two independent cultures. (C) Decrease in ROS accumulation associated with ahpC deficiency. Peroxide levels in dnaB-Ts (strain 2429) anddnaB-Ts–ΔahpC (strain 3780) during incubation at 42 °C were measured using Peroxy Orange-1 and flow cytometry. n = 5. (D) Suppression of the hyperlethalphenotype due to a ΔkatG mutation in a dnaB-Ts mutant by an ahpC deficiency (strain 4503). n = 3. n, number of independent replicates. Data are mean ± SD.

4 of 8 | www.pnas.org/cgi/doi/10.1073/pnas.1901730116 Hong et al.

Dow

nloa

ded

by g

uest

on

June

12,

202

0

(SI Appendix, Fig. S9C). These data suggest that the sublethalperoxide kills E. coli by adding to the ROS stimulated by ampi-cillin. As a control, we found that addition of a sublethal level ofperoxide to cultures stressed with a bacteriostatic agent (chlor-amphenicol) showed little killing (SI Appendix, Fig. S9D).Collectively, our peroxide data support a role for ROS in stress-

mediated lethality and show that if high levels of ROS are achievedin broth, cells die before stressor removal and plating. Once cellsare dead, subsequent reduction in ROS with antioxidants cannotrevive them. Thus, ROS present before stressor removal may maskthe detection of subsequent ROS-mediated poststress killing onagar (SI Appendix, Fig. S2 I and J).

DiscussionThe work described above supports the idea that diverse stressors(nalidixic acid, trimethoprim, ampicillin, and thermal stress with adnaB-Ts mutant) create potentially lethal primary lesions and alsostimulate a self-amplifying accumulation of toxic ROS. The lattercan be essential for cell death, as indicated by the lethal action ofseveral diverse stressors being halted or reduced by treatment withanti-ROS agents even after removal of the primary stressor (Fig. 1

B–D, F, andG and SI Appendix, Figs. S1G, S2 E, I, and J, and S4H).As expected, ROS continued to increase inside cells after removalof stressors (Fig. 1E and SI Appendix, Fig. S3). An example wasfound in which a primary antimicrobial lesion was necessary butinsufficient for bacterial death; when cells were treated with nali-dixic acid and ROS accumulation was suppressed, death did notoccur unless cells were deficient in the repair of quinolone-generatedlesions (Figs. 3 A and B and 4 A and B and SI Appendix, Figs. S1 Eand F and S6 D and E). A similar phenomenon was observed forkanamycin-mediated killing and hslV-mediated repair of proteindamage (Fig. 4 C and D). Since ROS have many toxic effects (46),including DNA breakage (11, 26, 47), membrane depolarization(48) (SI Appendix, Fig. S10), protein carbonylation (49), and cellwall damage (Fig. 5 C and D), ROS-mediated damage likely trig-gers subsequent rounds of ROS production, leading to even moredamage. Accumulated damage eventually overwhelms the capac-ity of a cell to repair potentially lethal primary lesions and ensurescell death. Our current view of stress-mediated lethality, whichincludes both direct lesion-based death and ROS-based death, issketched in Fig. 6.

D

0 h 0.5 h 1 h 1.5 h 2 h 2.5 h 3 h

30% 52% 61% 72% 97%~1%~1%

dnaB-TsC

5 μm

dnaB-Ts + bipyridyl + thiourea0 h

5 μm

2 h 3 h

5% 7% 7%~1%

E0 h 2 h 3 hdnaB-Ts ahpC

<5%

5 μm

~1% 5%

2.5 h

0 h 0.5 h 1 h 2 hNalidixic acidA Nalidixic acid + bipyridyl + thiourea

0 h 1 h 2 hB

5 μm 5 μm

~1% 70% 97% >98% 94%81%~1%

Fig. 3. Primary and ROS-mediated DNA damage. (A) DNA DSBs due to nalidixic acid (1 MIC, 8 μg/mL). DSBs in strain 4245 were detected by fluorescence mi-croscopy using RecN-YFP fusion. n = 4. (B) Non–ROS-mediated DSBs due to nalidixic acid treatment. As in A, thiourea (0.75 MIC), bipyridyl (0.5 MIC) were addedwith drug. n = 4. (C) DSB accumulation during 42 °C treatment (for indicated times) of dnaB-Ts strain 2429, detected as in A. n = 6. (D) Chemical suppression ofROS accumulation inhibits generation of DSBs during thermal stress with dnaB-Ts mutant. As in A, thiourea (0.5 MIC) and bipyridyl (0.5 MIC) were added tosuppress ROS accumulation. n = 3. (E) ΔahpC inhibits accumulation of DSBs during thermal stress of dnaB-Ts strain 3780 as in A. n = 4. In all panels, values of nindicate the number of independent experiments, arrows indicate DSB fluorescent foci, and numbers below panels indicate the percentage of cells with at leastone fluorescent focus. (Upper) Bright field images. (Lower) Fluorescent images.

Hong et al. PNAS Latest Articles | 5 of 8

MICRO

BIOLO

GY

Dow

nloa

ded

by g

uest

on

June

12,

202

0

While our data do not rule out small amounts of primary, stressor-specific damage persisting after stressor removal, such damage islikely to be of little consequence once ROS become self-amplifying.In cases where anti-ROS agents blocked killing (Figs. 1, 2, and 5),any persistent primary damage must be insufficient to kill cells. Wenote that the poststress death assay was performed shortly after theinitiation of antimicrobial treatment. If the assay is delayed, manycells die before drug removal and plating, thereby reducing theability of anti-ROS agents to distinguish ROS-mediated poststressdeath from overall killing. After long exposure to stressor, survivalcurves with anti-ROS agents in agar approach those obtained in theabsence of the agents (SI Appendix, Figs. S1G and S2E). In suchsituations, failure to detect ROS-mediated poststress death does notmean that ROS made no contribution to death, as ROS action likelyoccurred before plating on agar.We encountered two situations in which little effect of ROS-

mitigating agents was observed after washing, diluting, and platingcells. In the case of ampicillin (Fig. 1D), coadministration ofbipyridyl and/or thiourea with the antimicrobial inhibits killing (9,10), indicating that ROS contribute to ampicillin lethality. Killing viaROS apparently occurs rapidly, since observing poststress deathrequired cotreatment of broth cultures with anti-ROS agents (Fig.

1D and SI Appendix, Fig. S2 I and J), presumably to suppress rapidROS accumulation and death before drug removal and plating.With kanamycin treatment (Fig. 4 C and D), coadministration ofbipyridyl and/or thiourea with the antimicrobial inhibits killing (9,10), indicating that ROS contribute to cell death. Unlike the situa-tion with ampicillin, cotreatment of kanamycin with bipyridyl plusthiourea before plating did not reveal poststress killing (SI Ap-pendix, Fig. S10H). When we measured residual drug by massspectrometry following washing and dilution of cells, we found thatkanamycin was not effectively removed; cell-bound kanamycin de-creased by only 50% (SI Appendix, Fig. S11), much less than thedegree observed with other antimicrobials (SI Appendix, Figs. S1H andS2 F and K). This carryover of kanamycin precludes measurement ofkanamycin-mediated poststress ROS-mediated killing using thewashing/dilution method. Overall, our data emphasize that detectionof poststress ROS effects requires effective residual drug removal andmay require damping of ROS-mediated toxicity before plating.How initial lesions lead to ROS accumulation is poorly un-

derstood. Since many types of stressor-specific damage occur,multiple pathways likely exist. Evidence for distinct processes is seenin a comparison of genes involved in death stimulated by quino-lones and thymine starvation (genes participating in downstream

0 h 2.

5 h% In

put c

ells

reco

vere

d A dnaB-Ts Nalidixic acid

0 h

2 h

Aga

r + B

ip_2 h

H 2O 2

20 m

in_2 h

B102

100

10-2

102

100

0 h 1.5 h3 hAmpAmp

Bip-Thi+

Control 5 min

Bip-Thi + Amp_2.5 h

1 h

H2O2

Bip-Thi+

H2O2

+

C D

10-2

H 2O 2

_Aga

r + B

ip_2 h

Aga

r + Thi_

2.5 h

H 2O 2

20 m

in_2.5

h

H 2O 2

_Aga

r + Thi_

2.5 h

Fig. 5. ROS-mediated exacerbation of stress-stimulated death. (A) Exogenous peroxide-stimulated killing of dnaB-Ts mutant. Strain 2429 was shifted to 42 °C for130 min and then treated for 20 min with 1.5 mM H2O2, a sublethal dose (SI Appendix, Fig. S9), before shifting to 30 °C and plating on agar lacking/containingthiourea (Thi) at 15 mM (0.15 MIC). n = 4. Data are mean ± SD. (B) Peroxide-stimulated killing by nalidixic acid (8 μg/mL (1 MIC) for 100 min, followed by 20 min of1.5 mM H2O2). After drug and peroxide treatments, cells were plated on agar lacking/containing bipyridyl (Bip) at 0.5 MIC. n = 5. Data are mean ± SD. (C) In-hibition of ampicillin (2.5 MIC)-mediated lysis of strain 3001 by bipyridyl (0.4 MIC) plus thiourea (0.45 MIC). n = 4. (D) Peroxide-stimulated, ampicillin-mediated celllysis. Following cotreatment by ampicillin (Amp), bipyridyl and thiourea (concentrations as in C) for 2.5 h, cells were resuspended in fresh medium and then treatedwith 1.5 mM H2O2 with/without bipyridyl and thiourea for the indicated times (n = 4). n, number of independent replicates; all replicates gave similar results.

A B

% In

put c

ells

reco

vere

d

0 1 2Time (h)

Wild-type

102

100

10-2

recA

3

recABC

0 1 2Time (h)

3

Nalidixic + Bip + Thi Nalidixic acid

102

100

10-4

10-2

0 1 2Time (h)

CKanamycin + Bip + Thi

0 1 2Time (h)

Wild-type

D Kanamycin

ΔΔ hslVΔ

Fig. 4. Deficiency in primary damage repair reduces dependence on ROS for killing by quinolones and aminoglycosides. (A) Reduced dependence on ROS forkilling by nalidixic acid in recA and recABCDNA damage repair mutants. Midlog phase cultures of wild-type strain (strain 10), recA-13 (strain 14), and recA-13 recB-21 recC-22 (strain 156) were treated with bipyridyl (Bip, 0.5 MIC) plus thiourea (Thi, 0.5 MIC) and nalidixic acid (10 MIC) for the indicated times before dilution,plating, and determination of cfu. (B) Killing by nalidixic acid increased by rec mutations. Conditions and strains were as in A, except that bipyridyl and thioureawere omitted. (C) Reduced dependence on ROS for killing by kanamycin with the hslV protein damage repair mutant. Midlog phase cultures of wild-type strain(3001) and ΔhslV (strain 3325) were treated with kanamycin (1.1 MIC) in the presence of bipyridyl (0.5 MIC) plus thiourea (0.5 MIC) for the indicated times beforedilution, plating, and determination of cfu. (D) Killing by kanamycin increased by protein damage repair mutation. Strains and conditions were as in C, except thatbipyridyl and thiourea were omitted. In A and C, the addition of antioxidants had no effect on antimicrobial MIC. In all panels, n = 3. Data are mean ± SD.

6 of 8 | www.pnas.org/cgi/doi/10.1073/pnas.1901730116 Hong et al.

Dow

nloa

ded

by g

uest

on

June

12,

202

0

events are identified by allelic deficiencies that are protective fromlethal stress). With quinolones, such protective deficiencies occurin mazF, lepA, cpx, and arcA (12, 50, 51); in contrast, thyminelessdeath, which also involves ROS accumulation, is not influenced bymutations in these genes (26). Genes involved in iron metabolism,the TCA cycle, and oxidative stress have also been implicated inROS production and accumulation for a variety of stressors (14,21). Learning how the effects of diverse types of stressor convergeto induce an ROS cascade will require more work.The existence of a toxic, self-amplifying process implies that

protective mechanisms exist for survival in a variety of environ-ments. Indeed, regulons that suppress the effects of oxidative stresshave been studied for many years (52). In the present work, weobserved a protective effect arising from a deficiency in the ahpCFperoxidase genes in the dnaB-Ts background; the deficiency ele-vated the expression of many other oxidative stress-mitigating genesat permissive temperature, thereby suppressing ROS-mediateddeath that would otherwise follow exposure to restrictive temper-ature or to antimicrobials (Fig. 2 and SI Appendix, Fig. S7 and TableS2). A similar phenomenon likely follows a short treatment of E.coli with hydrogen peroxide (13) or treatment with low concentra-tions of paraquat or plumbagin (53). The latter stimulators of su-peroxide production protect from killing by bleomycin (54) andquinolones (53, 55); in contrast, high concentrations are toxic (56,57). This dichotomy suggests that superoxide is bifunctional: pro-tective at low concentration and destructive at high levels. Theprotective function likely derives from low to moderate levels of

superoxide and hydrogen peroxide inducing the expression of manyprotective oxidative stress-responsive genes (13).The existence of a runaway (i.e., self-amplifying) death mechanism

raises questions about the potential roles of ROS in developmentalprocesses sometimes termed programmed cell death. ROS amplifi-cation could serve as the terminal stage. ROS could also be involvedin the association of apoptotic biomarkers with stress-mediatedbacterial death (48, 58–61). For biomarkers, a relevant question iswhether the presence of a biomarker reflects death derived from aself-sustaining process or from a continuing stimulation of toxicevents by the initiating stressor. To address such questions, exam-ining the time course of events is important to establish that theobserved signals reflect a death pathway rather than a breakdownproduct of dead cells.Finally, the present work addresses challenges to the idea that

ROS contribute to killing from diverse antimicrobials (22–24).The ability of ROS-mitigating chemicals and especially of cata-lase to block death after stressor removal (Fig. 1) provides strongevidence for a causal role of ROS in antimicrobial-mediatedkilling. It has been argued that potential off-target effects are dif-ficult to rule out for perturbations with chemical agents, such asthiourea and bipyridyl (23, 24); thus, statements of causality arequestionable when such agents are used. Off-target effects are un-likely with enzymatic suppression of ROS, as when catalase is addedto agar or when a deficiency of katG (catalase) increases killing (20,26, 28). A causal role for ROS in stress-mediated death has alsoemerged from the effects of mutations in a variety of other genes,most notably genes involved in the processing of 8-oxoguanine in

DNA damage

Antioxidant agent

(iii)

ROS

Additional damage

Cell death

(v)(vi)

Antioxidant gene

DNA replication

Fork stalling

ROS

Membrane depolarizationDNA breaks

Other damage ?

(i)

(ii)

dnaB-Ts

(viii)

Survival

(iv)

(vii)

Primary damage

(i)

(iv)

Nalidixic acidTrimethoprim

Cell wall

Ampicillin(ii)

DNA

Antioxidant agent

ROS

Secondary damage

Cell death

Antioxidant gene

Membrane depolarization

DNA breaks

Other damage ?

(vii)

Survival

(v)

(vi)

Iron chelatorSlow respiration

Wall damage

Protein carbonylationLipid peroxidation

Ribosome inhibitor

A B(iii)

Aminoglycoside

Protein

Repair(viii)

Repair

Fig. 6. Schemes describing poststress death and relationship to direct killing by stress. (A) dnaB-Ts–mediated poststress death. Inactivation of the DnaBhelicase by incubation of the dnaB-Ts mutant at nonpermissive temperature stalls replication forks (i), which stimulates ROS accumulation by unknownmechanisms (ii). ROS generate early DNA damage that is insufficient to directly cause death (iii). However, the DNA damage likely stimulates further ROSaccumulation (iv). ROS attack macromolecules, causing secondary damage, such as DNA breakage, protein carbonylation, and membrane depolarization (v).Secondary damage is expected to further elevate levels of ROS (vi). Extensive damage is expected to kill cells (vii). Reduction of ROS, as seen with deficiency ofahpCF or with antioxidants (catalase, thiourea), leads to cell survival (viii). (B) Poststress death and direct killing due to chemical and physical stressors. Soonafter application of stress, DNA damaging agents (i), β-lactams (ii), and aminoglycosides (iii) create primary damage. Such damage, if severe or within aprimary repair deficiency background, may directly kill cells without the need for ROS (iv). Primary damage, which can be insufficient to kill cells, activatespathways leading to the accumulation of ROS (v), which can then cause secondary damage and more ROS accumulation. When damage is too great to berepaired (vi), it causes death. ROS accumulation can be blocked or reduced in several ways, such as by overexpression of antioxidant genes, the addition ofantioxidants or iron chelators, and inhibition of respiration (vii). Such blockage can lead to survival, which is called tolerance. ROS may contribute to killing byoverwhelming repair of primary stress damage (viii), since deficiencies in primary damage repair reduce the dependence on ROS to kill cells.

Hong et al. PNAS Latest Articles | 7 of 8

MICRO

BIOLO

GY

Dow

nloa

ded

by g

uest

on

June

12,

202

0

DNA (reviewed in ref. 30). Whether endogenous ROS levels arehigh enough to kill bacterial cells (24) has been addressed in part bystressors creating lesions that are hypersensitive to ROS attack (20,26) and, in the present work, by self-amplification of ROS. Weconclude that ROS contribute to the lethal action of many stressortypes. A next step is to find adjunctive therapies for antimicrobialsto more rapidly cause ROS to become self-amplifying.

Materials and MethodsDetailed descriptions of the construction of E. coli strains, poststress survival assayfor dnaB-Ts and antibiotic-treated cells, measurement of ROS and membrane

depolarization using flow cytometry, detection of DNA damage using fluores-cent microscopy, assay of protein carbonylation and lipid peroxidation, de-termination of drug concentrations using LC/MS-MS, and RNA-seq methodologyare provided in SI Appendix. The strains used in this study are listed in SI Ap-pendix, Table S3.

ACKNOWLEDGMENTS. We thank Marila Gennaro, Bo Shopsin, and severalanonymous reviewers for critical comments; Chao Chen, Yan Pan, andVeronique Dartois for help with the LC/MS-MS analysis; and Sanjay Tyagi forhelp with microscopy. This work was supported by grants from the USNational Institutes of Health (DP2OD007423 and R01 AI073491) and theChina National Natural Science Foundation (81661138005 and 81473251).

1. Ventola CL (2015) The antibiotic resistance crisis, part 1: Causes and threats. PT 40:277–283.

2. Malik M, Hoatam G, Chavda K, Kerns RJ, Drlica K (2010) Novel approach for com-paring the abilities of quinolones to restrict the emergence of resistant mutantsduring quinolone exposure. Antimicrob Agents Chemother 54:149–156.

3. Riesenfeld C, Everett M, Piddock LJ, Hall BG (1997) Adaptive mutations produce re-sistance to ciprofloxacin. Antimicrob Agents Chemother 41:2059–2060.

4. Baym M, et al. (2016) Spatiotemporal microbial evolution on antibiotic landscapes.Science 353:1147–1151.

5. Liu Y, et al. (2016) Resveratrol antagonizes antimicrobial lethality and stimulates re-covery of bacterial mutants. PLoS One 11:e0153023.

6. McKenzie GJ, Harris RS, Lee PL, Rosenberg SM (2000) The SOS response regulatesadaptive mutation. Proc Natl Acad Sci USA 97:6646–6651.

7. Hastings PJ, Rosenberg SM, Slack A (2004) Antibiotic-induced lateral transfer of an-tibiotic resistance. Trends Microbiol 12:401–404.

8. Cirz RT, et al. (2005) Inhibition of mutation and combating the evolution of antibioticresistance. PLoS Biol 3:e176.

9. Kohanski MA, Dwyer DJ, Hayete B, Lawrence CA, Collins JJ (2007) A common mech-anism of cellular death induced by bactericidal antibiotics. Cell 130:797–810.

10. Wang X, Zhao X (2009) Contribution of oxidative damage to antimicrobial lethality.Antimicrob Agents Chemother 53:1395–1402.

11. Foti JJ, Devadoss B, Winkler JA, Collins JJ, Walker GC (2012) Oxidation of the guaninenucleotide pool underlies cell death by bactericidal antibiotics. Science 336:315–319.

12. Dorsey-Oresto A, et al. (2013) YihE kinase is a central regulator of programmed celldeath in bacteria. Cell Rep 3:528–537.

13. Dwyer DJ, et al. (2014) Antibiotics induce redox-related physiological alterations aspart of their lethality. Proc Natl Acad Sci USA 111:E2100–E2109.

14. Dwyer DJ, Collins JJ, Walker GC (2015) Unraveling the physiological complexities ofantibiotic lethality. Annu Rev Pharmacol Toxicol 55:313–332.

15. Zhao X, Drlica K (2014) Reactive oxygen species and the bacterial response to lethalstress. Curr Opin Microbiol 21:1–6.

16. Van Acker H, Coenye T (2017) The role of reactive oxygen species in antibiotic-mediated killing of bacteria. Trends Microbiol 25:456–466.

17. Mironov A, et al. (2017) Mechanism of H2S-mediated protection against oxidativestress in Escherichia coli. Proc Natl Acad Sci USA 114:6022–6027.

18. Shatalin K, Shatalina E, Mironov A, Nudler E (2011) H2S: A universal defense againstantibiotics in bacteria. Science 334:986–990.

19. Gusarov I, Shatalin K, Starodubtseva M, Nudler E (2009) Endogenous nitric oxideprotects bacteria against a wide spectrum of antibiotics. Science 325:1380–1384.

20. Luan G, Hong Y, Drlica K, Zhao X (2018) Suppression of reactive oxygen species ac-cumulation accounts for paradoxical bacterial survival at high quinolone concentra-tion. Antimicrob Agents Chemother 62:e01622-17.

21. Brynildsen MP, Winkler JA, Spina CS, MacDonald IC, Collins JJ (2013) Potentiatingantibacterial activity by predictably enhancing endogenous microbial ROS pro-duction. Nat Biotechnol 31:160–165.

22. Keren I, Wu Y, Inocencio J, Mulcahy LR, Lewis K (2013) Killing by bactericidal anti-biotics does not depend on reactive oxygen species. Science 339:1213–1216.

23. Liu Y, Imlay JA (2013) Cell death from antibiotics without the involvement of reactiveoxygen species. Science 339:1210–1213.

24. Imlay JA (2015) Diagnosing oxidative stress in bacteria: Not as easy as you mightthink. Curr Opin Microbiol 24:124–131.

25. Zhao X, Hong Y, Drlica K (2015) Moving forward with reactive oxygen species in-volvement in antimicrobial lethality. J Antimicrob Chemother 70:639–642.

26. Hong Y, Li L, Luan G, Drlica K, Zhao X (2017) Contribution of reactive oxygen speciesto thymineless death in Escherichia coli. Nat Microbiol 2:1667–1675.

27. Dong TG, et al. (2015) Generation of reactive oxygen species by lethal attacks fromcompeting microbes. Proc Natl Acad Sci USA 112:2181–2186.

28. Takahashi N, et al. (2017) Lethality of MalE-LacZ hybrid protein shares mechanisticattributes with oxidative component of antibiotic lethality. Proc Natl Acad Sci USA114:9164–9169.

29. Fan XY, et al. (2018) Oxidation of dCTP contributes to antibiotic lethality instationary-phase mycobacteria. Proc Natl Acad Sci USA 115:2210–2215.

30. Gruber CC, Walker GC (2018) Incomplete base excision repair contributes to cell deathfrom antibiotics and other stresses. DNA Repair (Amst) 71:108–117.

31. Goss WA, Deitz WH, Cook TM (1965) Mechanism of action of nalidixic acid on Es-cherichia coli, II: Inhibition of deoxyribonucleic acid synthesis. J Bacteriol 89:1068–1074.

32. Mustaev A, et al. (2014) Fluoroquinolone-gyrase-DNA complexes: Two modes of drugbinding. J Biol Chem 289:12300–12312.

33. Kantor GJ, Deering RA (1968) Effect of nalidixic acid and hydroxyurea on divisionability of Escherichia coli fil+ and lon- strains. J Bacteriol 95:520–530.

34. Snyder M, Drlica K (1979) DNA gyrase on the bacterial chromosome: DNA cleavageinduced by oxolinic acid. J Mol Biol 131:287–302.

35. Chow RT, Dougherty TJ, Fraimow HS, Bellin EY, Miller MH (1988) Association betweenearly inhibition of DNA synthesis and the MICs and MBCs of carboxyquinolone anti-microbial agents for wild-type and mutant [gyrA nfxB(ompF) acrA] Escherichia coli K-12. Antimicrob Agents Chemother 32:1113–1118.

36. Aldred KJ, et al. (2014) Role of the water-metal ion bridge in mediating interactionsbetween quinolones and Escherichia coli topoisomerase IV. Biochemistry 53:5558–5567.

37. Giroux X, Su W-L, Bredeche M-F, Matic I (2017) Maladaptive DNA repair is the ulti-mate contributor to the death of trimethoprim-treated cells under aerobic and an-aerobic conditions. Proc Natl Acad Sci USA 114:11512–11517.

38. Yao Z, Kahne D, Kishony R (2012) Distinct single-cell morphological dynamics underbeta-lactam antibiotics. Mol Cell 48:705–712.

39. Storz G, et al. (1989) An alkyl hydroperoxide reductase induced by oxidative stress inSalmonella typhimurium and Escherichia coli: Genetic characterization and cloning ofahp. J Bacteriol 171:2049–2055.

40. Hong Y, Zhao X (2018) RNA-seq analysis of transcriptomes of dnaB-Ts and dnaB-TsΔahpC at both permissive and non-permissive temperature. Gene Expression Omni-bus. Available at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE114262.Deposited May 9, 2018.

41. Kidane D, Sanchez H, Alonso JC, Graumann PL (2004) Visualization of DNA double-strand break repair in live bacteria reveals dynamic recruitment of Bacillus subtilisRecF, RecO, and RecN proteins to distinct sites on the nucleoids. Mol Microbiol 52:1627–1639.

42. Chen CR, Malik M, Snyder M, Drlica K (1996) DNA gyrase and topoisomerase IV on thebacterial chromosome: Quinolone-induced DNA cleavage. J Mol Biol 258:627–637.

43. Malik M, Zhao X, Drlica K (2006) Lethal fragmentation of bacterial chromosomesmediated by DNA gyrase and quinolones. Mol Microbiol 61:810–825.

44. Krause KM, Serio AW, Kane TR, Connolly LE (2016) Aminoglycosides: An overview.Cold Spring Harb Perspect Med 6:1–18.

45. Poirel L, Jayol A, Nordmann P (2017) Polymyxins: Antibacterial activity, susceptibilitytesting, and resistance mechanisms encoded by plasmids or chromosomes. ClinMicrobiol Rev 30:557–596.

46. Imlay JA (2003) Pathways of oxidative damage. Annu Rev Microbiol 57:395–418.47. Kobayashi S, Ueda K, Komano T (1990) The effects of metal ions on the DNA damage

induced by hydrogen peroxide. Agric Biol Chem 54:69–76.48. Dwyer DJ, Camacho DM, Kohanski MA, Callura JM, Collins JJ (2012) Antibiotic-

induced bacterial cell death exhibits physiological and biochemical hallmarks of ap-optosis. Mol Cell 46:561–572.

49. Dalle-Donne I, Rossi R, Giustarini D, Milzani A, Colombo R (2003) Protein carbonylgroups as biomarkers of oxidative stress. Clin Chim Acta 329:23–38.

50. Li L, et al. (2014) Ribosomal elongation factor 4 promotes cell death associated withlethal stress. MBio 5:e01708.

51. Kohanski MA, Dwyer DJ, Wierzbowski J, Cottarel G, Collins JJ (2008) Mistranslation ofmembrane proteins and two-component system activation trigger antibiotic-mediated cell death. Cell 135:679–690.

52. Farr SB, Kogoma T (1991) Oxidative stress responses in Escherichia coli and Salmonellatyphimurium. Microbiol Rev 55:561–585.

53. Mosel M, Li L, Drlica K, Zhao X (2013) Superoxide-mediated protection of Escherichiacoli from antimicrobials. Antimicrob Agents Chemother 57:5755–5759.

54. Burger RM, Drlica K (2009) Superoxide protects Escherichia coli from bleomycin-mediated lethality. J Inorg Biochem 103:1273–1277.

55. Wu Y, Vuli�c M, Keren I, Lewis K (2012) Role of oxidative stress in persister tolerance.Antimicrob Agents Chemother 56:4922–4926.

56. Hassan HM, Fridovich I (1978) Superoxide radical and the oxygen enhancement of thetoxicity of paraquat in Escherichia coli. J Biol Chem 253:8143–8148.

57. Farr SB, Natvig DO, Kogoma T (1985) Toxicity and mutagenicity of plumbagin and theinduction of a possible new DNA repair pathway in Escherichia coli. J Bacteriol 164:1309–1316.

58. Bos J, Yakhnina AA, Gitai Z (2012) BapE DNA endonuclease induces an apoptotic-likeresponse to DNA damage in Caulobacter. Proc Natl Acad Sci USA 109:18096–18101.

59. Wadhawan S, Gautam S, Sharma A (2010) Metabolic stress-induced programmed celldeath in Xanthomonas. FEMS Microbiol Lett 312:176–183.

60. Gautam S, Sharma A (2002) Rapid cell death in Xanthomonas campestris pv. glycines.J Gen Appl Microbiol 48:67–76.

61. Dewachter L, et al. (2015) A single amino acid substitution in Obg activates a newprogrammed cell death pathway in Escherichia coli. MBio 6:e01935-15.

8 of 8 | www.pnas.org/cgi/doi/10.1073/pnas.1901730116 Hong et al.

Dow

nloa

ded

by g

uest

on

June

12,

202

0